Nvidia tops TSMC's queue while AMD noses forward

Morgan Stanley said Nvidia will remain TSMC's largest CoWoS customer in 2027 while AMD's EPYC Venice platform could surpass Nvidia's Vera CPU shipments that year. The firm projects TSMC will reach wafer capacity of 200,000 wafers per month in 2027. Nvidia uses TSMC's CoWoS-L packaging for AI GPUs such as Blackwell and Rubin and CoWoS-R for Vera CPUs. CoWoS-L capacity is expected to hit about 910,000 units, up 40 percent year on year. Vera shipments are expected to double, helping lift Nvidia's data center revenue by 52 percent. Nvidia has delivered first Vera CPUs to Anthropic, OpenAI, SpaceX and Oracle. AMD's Venice platform is already in volume production at TSMC. Based on Zen 6 architecture, Venice targets AI and high-performance computing. Morgan Stanley projects Vera could reach 5.75 million units in 2027 while Venice may reach 6.75 million units, 17 percent more than Vera and 5.4 times expected 2026 volume. Venice uses TSMC's 2-nanometer process while Vera uses 3-nanometer. Custom silicon from cloud providers including OpenAI, Google and Amazon presents a challenge for both companies.